ASD Essential Eight Maturity Level 2

Discovers every reachable application and fingerprints versions across the estate — including bespoke, line-of-business and unsupported releases — and exploits known CVEs as part of attack chains, surfacing where patching has slipped against the ML1 and ML2 windows.

Continuously discovers internet-facing services and exploits unpatched vulnerabilities the way a ransomware actor would, surfacing on every scheduled run where internet-facing online services remain exploitable past the 48-hour critical and two-week non-critical patching windows — evidence that patching has slipped, not proof the cadence was met.

Horizon3.ai's attack team publishes targeted tests for newly disclosed critical CVEs — often within hours of disclosure, and in some cases ahead of the public patch — so exposure to the 48-hour window at ML1 (online services) and ML3 (productivity software) is auditable in the window between a CVE going public and your next scan. The catalogue extends beyond network appliances to Microsoft and Linux-hosted software — SharePoint and CyberPanel among the published examples.

Ingests existing scanner exports across every application class and ranks CVE-asset pairs by exploitability — so triage inside the ML2 one-month and ML3 two-week windows is governed by real attack reach rather than CVSS alone.

Discovers every reachable host, fingerprints OS version and build, and exploits missing patches as part of lateral-movement chains — direct evidence of unpatched workstations, internal servers and unsupported OS releases that should have been replaced.

Discovers and exploits unpatched internet-facing servers and network devices, providing attack-side evidence on every scheduled run of where perimeter OS patching remains exploitable past the 48-hour critical and two-week non-critical windows — evidence patching has slipped, not proof the cadence was met.

Targeted tests for newly disclosed critical OS CVEs — both at the perimeter (ML1 internet-facing) and inside the network (ML3 workstations and non-internet servers) — let you verify exposure in the window between a CVE going public and your next scan. The catalogue includes Microsoft server components — a WSUS misconfiguration test card among recent additions — alongside the appliance CVEs it is best known for.

Re-ranks workstation and non-internet OS CVE-asset pairs by exploitability, classifying them as Confirmed Exploitable, Contextually Exploitable or Threat Actor Pressure — attacker-first prioritisation inside the ML1 one-month and ML3 one-month windows.

Discovers and tests every authentication endpoint reached during an internal pentest, surfacing privileged and unprivileged accounts not actually protected by MFA, and reaching databases, file shares and source-code repositories where MFA on data-repository access (ML3) hasn't been enforced.

Supports credential injection — inject a compromised credential into an external pentest and NodeZero runs the test from an authenticated perspective, proving the real impact of a stolen credential from outside — attack-impact evidence toward the MFA obligations, not a live test of the deployed factor itself.

Takes the credentials your phishing tool captures and demonstrates what an attacker actually reaches across internal, third-party and customer services — attack-impact evidence of where MFA is absent or fails to contain a phished credential. It does not run a live relay against the factor, so it does not on its own prove SMS, push or FIDO2 meets the phishing-resistant qualifier ML2 introduces and ML3 strengthens.

NodeZero's WebApp Pentesting tests reachable customer-facing and third-party authentication surfaces for exploitable weaknesses where they form real attack paths that defeat the factor protecting sensitive customer data.

Surfaces the weak, breached and reused passwords MFA is expected to compensate for — direct evidence of the failure mode if the factor itself is bypassed.

Attacks the Entra identity plane — including Seamless SSO Silver Ticket and Entra Connect credential dumping — proving whether MFA on privileged users actually resists documented bypass techniques.

Starts from a real authenticated user's credentials and proves what that user can actually reach — attack-impact evidence of where a single credential is all that stands between a user and the systems and data repositories MFA is meant to gate.

Autonomous decoy AD accounts baited for Kerberoasting and AS-REP roasting, paired with DC Kerberos logging and real-time SIEM/SOAR alerting — validates whether a live credential-theft attempt against the identity plane is actually detected, the detection half of the ML2/ML3 logging-and-analysis obligation below. Central log collection, log protection and incident-response enactment remain outside autonomous testing.

Proves end-to-end attack paths from unprivileged access to domain compromise, and attempts LSASS dumping and NTLM-hash extraction from host memory — successful extraction is direct evidence of separation, LSA Protection, Credential Guard or Remote Credential Guard failures across ML1 and ML3. Kerberoasting and AS-REP roasting are separate identity attacks against the KDC, not evidence about Credential Guard's state.

Simulated phishing compromise of a privileged account proves whether the ML1 internet/email isolation control actually prevents that account being abused after credential theft.

AD attack-graph visualisation makes the cross-tier logon paths that break the ML1 unprivileged/privileged separation control visible and fixable.

Starts from a real privileged or unprivileged user perspective and proves what they can actually reach — empirical evidence of the ML1 separation rule and the ML3 least-privilege limitation on systems, applications and data repositories.

First AI to solve GOAD in 14 minutes (Horizon3-reported); it is not uncommon for NodeZero to crack more than 50% of the passwords it tests on a first audit. Audits break-glass, local-administrator and service-account credentials against the ML2 long, unique and managed requirement, and surfaces weak or never-rotated privileged accounts worth reviewing against the ML2 12-month revalidation and 45-day inactivity thresholds — the lifecycle disable decision itself remains an identity-governance check.

Hybrid and cloud environments routinely break the ML2 rule that privileged operating environments not be virtualised within unprivileged ones; cloud pentesting tests privilege-escalation paths between on-prem and AWS, Azure or Kubernetes.

Autonomously infers crown-jewel systems and identities — executive accounts, production and virtualisation infrastructure — and proves the attack paths that reach them: additive attack-impact evidence for the ML3 least-privilege rule, on top of what Insider Threat Testing already demonstrates, that an over-provisioned account's reach actually extends to crown jewels.

Supports credential injection — inject a compromised credential into an external pentest and NodeZero proves from outside what that account actually reaches across the internet-facing estate, attack-impact evidence of whether a stolen privileged credential is contained.

Autonomous decoy AD accounts baited for Kerberoasting and AS-REP roasting, paired with DC Kerberos logging and real-time SIEM/SOAR alerting — validates whether a live identity attack against a privileged account actually surfaces to the SOC, the detection half of the ML2/ML3 logging-and-analysis obligation below. Central log collection, log protection and incident-response enactment remain outside autonomous testing.

Per-host block/allow telemetry across 40+ EDR vendors surfaces workstations (ML1) and non-internet servers (ML3) where successful payload execution proves neither EDR nor application-control policy intervened — the exact gap each tier of application-control implementation is meant to close.

Drops and executes payloads from %TEMP%, %APPDATA%, browser cache, system directories and ProgramData paths during attack chains, attempts BYOVD (Bring Your Own Vulnerable Driver) and known LOLBins — successful execution is direct evidence the ML1, ML2 and ML3 application-control rulesets aren't enforcing.

Attempts execution against internet-facing servers from the attacker perspective — proves whether the ML2 application-control implementation is actually present and enforcing on the servers attackers touch first.

Autonomous decoy AD accounts baited for Kerberoasting and AS-REP roasting, paired with DC Kerberos logging and real-time SIEM/SOAR alerting — validates whether a live identity attack is actually detected, the detection half of the ML2/ML3 logging-and-analysis obligation below. Central log collection, log protection and incident-response enactment remain outside autonomous testing.

Takes the credentials your phishing tool captures and proves their blast radius inside the network — attack-impact evidence of the consequence the macro-restriction rules are meant to prevent. NodeZero does not deliver or detonate a macro, so whether the macro-execution restrictions enforce (internet-origin blocking, Win32-API blocking, trusted-source/sandbox restriction) sits alongside the other macro-policy items in the gap, not as direct evidence.

Fingerprints installed browsers, Java runtimes, .NET versions and PowerShell engines during host enumeration — surfaces the IE 11, browser-Java, .NET 3.5 and PowerShell 2.0 surfaces ML1 and ML3 require to be removed, and chains Office and PDF child-process and PowerShell Constrained Language Mode failures into attack paths.

Per-host EDR block/allow telemetry surfaces where generic attacker techniques are or aren't stopped on a host — supporting evidence around the ML2 Office-hardening rules. NodeZero does not detonate Office to trigger these specific ASR rules, so their enforcement is confirmed by configuration review.

Proves the blast radius of a phished credential inside the network — attack-impact evidence of the consequence the ML2 OLE-prevention rule is meant to prevent. NodeZero does not deliver or detonate OLE-package lures or .HTA droppers, so the rule's enforcement is a hardening-policy check (see gap).

Autonomous decoy AD accounts baited for Kerberoasting and AS-REP roasting, paired with DC Kerberos logging and real-time SIEM/SOAR alerting — validates whether a live identity attack is actually detected, the detection half of the ML2/ML3 logging-and-analysis obligation below. Central log collection, log protection and incident-response enactment remain outside autonomous testing.

Chains lateral-movement attack paths through backup storage from compromised unprivileged and privileged (non-backup-admin) accounts — proves whether the ML1, ML2 and ML3 access, modification and self-access restrictions actually hold against the way ransomware operators reach backups.

Places decoy AWS credential files, MySQL dumps and Windows process monitors near backup infrastructure and within retained-backup paths — silent until triggered, then alerts to your SIEM when a non-backup privileged or backup-admin account touches them, surfacing the ML2 and ML3 modification and deletion controls that should have prevented it.